The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection vulnerability surface found in the qsh-form template.
Ingestion points: User-supplied input from the task and attachments fields in SKILL.md is interpolated directly into the final prompt.
Boundary markers: Absent. The template does not use delimiters (such as XML tags or triple backticks) or instructions to treat the interpolated content as untrusted data.
Capability inventory: The skill allows the agent to execute shell commands and Python scripts (scripts/orthogonalize.py) that perform file system operations.
Sanitization: No sanitization or escaping is performed on the user-provided data before interpolation.
[DYNAMIC_EXECUTION]: The main execution script uses dynamic path manipulation to load dependencies.
Evidence: scripts/orthogonalize.py uses sys.path.insert to add a relative directory (skill-pandadata-api/scripts) to the Python search path to import the pandadata_runtime module.
[EXTERNAL_DOWNLOADS]: The skill instructions prompt the agent to perform automatic package installation.
Evidence: SKILL.md instructs the agent to detect missing environments and "automatically install" dependencies including numpy, pandas, scipy, and pyarrow using pip, uv, or conda. These are well-known services, but the automated nature without user confirmation is noted.

factor-orthogonalize